在传统研究中,关于如何形成最能模拟目标的欺骗干扰,以及如何提取最能辨识欺骗干扰的特征,都没有现成的理论方法。文中提出了解决这一问题的研究框架,引入深度卷积神经网络(CNN)作为判决器训练反欺骗干扰,再验证CNN网络的对抗性样本风险。在此基础上,以CNN网络为基础构建物理约束的生成对抗网络,从而实现欺骗干扰与反欺骗干扰闭环连接,通过数据训练使二者相互激励并趋向纳什均衡。

Traditionally, no available solution can be followed about how to bestly imitate targets with deception jamming and how to extract the optimal discriminant characteristics. Thus, a research framework is proposed to solve the problem. Firstly, the convolutional neural network (CNN) is introduced as discriminators to train anti-deception jamming. Secondly, the adversarial risks of CNN is verified. Based on that, physics-constrained generative adversarial network is constructed which connects deception jamming and anti-deception jamming into close-loops, and then through the network training the two counterparts reinforce each other and the game between them approaches to the Nash equilibrium.

TN973.3